socket.c@ 20297

Last change on this file since 20297 was 20297, checked in by vboxsync, 16 years ago
NAT: make port-forwarding the same speed as guest->remote
Property svn:eol-style set to `native`
File size: 31.2 KB

Line
1	/*
2	* Copyright (c) 1995 Danny Gasparovski.
3	*
4	* Please read the file COPYRIGHT for the
5	* terms and conditions of the copyright.
6	*/
7
8	#define WANT_SYS_IOCTL_H
9	#include <slirp.h>
10	#include "ip_icmp.h"
11	#include "main.h"
12	#ifdef __sun__
13	#include <sys/filio.h>
14	#endif
15	#if defined (RT_OS_WINDOWS)
16	#include <iphlpapi.h>
17	#include <icmpapi.h>
18	#endif
19
20
21	static void send_icmp_to_guest(PNATState, char , size_t, struct socket , const struct sockaddr_in *);
22	#ifdef RT_OS_WINDOWS
23	static void sorecvfrom_icmp_win(PNATState, struct socket *);
24	#else /* RT_OS_WINDOWS */
25	static void sorecvfrom_icmp_unix(PNATState, struct socket *);
26	#endif /* !RT_OS_WINDOWS */
27
28	void
29	so_init()
30	{
31	}
32
33
34	struct socket *
35	solookup(struct socket *head, struct in_addr laddr,
36	u_int lport, struct in_addr faddr, u_int fport)
37	{
38	struct socket *so;
39
40	for (so = head->so_next; so != head; so = so->so_next)
41	{
42	if ( so->so_lport == lport
43	&& so->so_laddr.s_addr == laddr.s_addr
44	&& so->so_faddr.s_addr == faddr.s_addr
45	&& so->so_fport == fport)
46	return so;
47	}
48
49	return (struct socket *)NULL;
50	}
51
52	/*
53	* Create a new socket, initialise the fields
54	* It is the responsibility of the caller to
55	* insque() it into the correct linked-list
56	*/
57	struct socket *
58	socreate()
59	{
60	struct socket *so;
61
62	so = (struct socket *)RTMemAllocZ(sizeof(struct socket));
63	if(so)
64	{
65	so->so_state = SS_NOFDREF;
66	so->s = -1;
67	#if !defined(RT_OS_WINDOWS)
68	so->so_poll_index = -1;
69	#endif
70	}
71	return so;
72	}
73
74	/*
75	* remque and free a socket, clobber cache
76	* VBOX_WITH_SLIRP_MT: before sofree queue should be locked, because
77	* in sofree we don't know from which queue item beeing removed.
78	*/
79	void
80	sofree(PNATState pData, struct socket *so)
81	{
82	struct socket *so_prev = NULL;
83	if (so == tcp_last_so)
84	tcp_last_so = &tcb;
85	else if (so == udp_last_so)
86	udp_last_so = &udb;
87
88	/* check if mbuf haven't been already freed */
89	if (so->so_m != NULL)
90	m_free(pData, so->so_m);
91	#ifndef VBOX_WITH_SLIRP_MT
92	if(so->so_next && so->so_prev)
93	{
94	remque(pData, so); /* crashes if so is not in a queue */
95	NSOCK_DEC();
96	}
97
98	RTMemFree(so);
99	#else
100	so->so_deleted = 1;
101	#endif
102	}
103
104	#ifdef VBOX_WITH_SLIRP_MT
105	void
106	soread_queue(PNATState pData, struct socket so, int ret)
107	{
108	*ret = soread(pData, so);
109	}
110	#endif
111
112	/*
113	* Read from so's socket into sb_snd, updating all relevant sbuf fields
114	* NOTE: This will only be called if it is select()ed for reading, so
115	* a read() of 0 (or less) means it's disconnected
116	*/
117	int
118	soread(PNATState pData, struct socket *so)
119	{
120	int n, nn, lss, total;
121	struct sbuf *sb = &so->so_snd;
122	size_t len = sb->sb_datalen - sb->sb_cc;
123	struct iovec iov[2];
124	int mss = so->so_tcpcb->t_maxseg;
125	QSOCKET_LOCK(tcb);
126	SOCKET_LOCK(so);
127	QSOCKET_UNLOCK(tcb);
128
129	DEBUG_CALL("soread");
130	DEBUG_ARG("so = %lx", (long )so);
131
132	/*
133	* No need to check if there's enough room to read.
134	* soread wouldn't have been called if there weren't
135	*/
136
137	len = sb->sb_datalen - sb->sb_cc;
138
139	iov[0].iov_base = sb->sb_wptr;
140	iov[1].iov_base = 0;
141	iov[1].iov_len = 0;
142	if (sb->sb_wptr < sb->sb_rptr)
143	{
144	iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
145	/* Should never succeed, but... */
146	if (iov[0].iov_len > len)
147	iov[0].iov_len = len;
148	if (iov[0].iov_len > mss)
149	iov[0].iov_len -= iov[0].iov_len%mss;
150	n = 1;
151	}
152	else
153	{
154	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
155	/* Should never succeed, but... */
156	if (iov[0].iov_len > len)
157	iov[0].iov_len = len;
158	len -= iov[0].iov_len;
159	if (len)
160	{
161	iov[1].iov_base = sb->sb_data;
162	iov[1].iov_len = sb->sb_rptr - sb->sb_data;
163	if(iov[1].iov_len > len)
164	iov[1].iov_len = len;
165	total = iov[0].iov_len + iov[1].iov_len;
166	if (total > mss)
167	{
168	lss = total % mss;
169	if (iov[1].iov_len > lss)
170	{
171	iov[1].iov_len -= lss;
172	n = 2;
173	}
174	else
175	{
176	lss -= iov[1].iov_len;
177	iov[0].iov_len -= lss;
178	n = 1;
179	}
180	}
181	else
182	n = 2;
183	}
184	else
185	{
186	if (iov[0].iov_len > mss)
187	iov[0].iov_len -= iov[0].iov_len%mss;
188	n = 1;
189	}
190	}
191
192	#ifdef HAVE_READV
193	nn = readv(so->s, (struct iovec *)iov, n);
194	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
195	#else
196	nn = recv(so->s, iov[0].iov_base, iov[0].iov_len, 0);
197	#endif
198	if (nn <= 0)
199	{
200	#if defined(RT_OS_WINDOWS)
201	/*
202	* Special case for WSAEnumNetworkEvents: If we receive 0 bytes that
203	* _could_ mean that the connection is closed. But we will receive an
204	* FD_CLOSE event later if the connection was _really_ closed. With
205	* www.youtube.com I see this very often. Closing the socket too early
206	* would be dangerous.
207	*/
208	int status, ignored;
209	unsigned long pending = 0;
210	status = WSAIoctl(so->s, FIONREAD, NULL, 0, &pending, sizeof(unsigned long), &ignored, NULL, NULL);
211	if (status < 0)
212	LogRel(("NAT:error in WSAIoctl: %d\n", WSAGetLastError()));
213	if (nn == 0 && (pending != 0))
214	{
215	SOCKET_UNLOCK(so);
216	return 0;
217	}
218	#endif
219	if (nn < 0 && (errno == EINTR \|\| errno == EAGAIN \|\| errno == EWOULDBLOCK))
220	{
221	SOCKET_UNLOCK(so);
222	return 0;
223	}
224	else
225	{
226	/* nn == 0 means peer has performed an orderly shutdown */
227	DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n",
228	nn, errno, strerror(errno)));
229	sofcantrcvmore(so);
230	tcp_sockclosed(pData, sototcpcb(so));
231	SOCKET_UNLOCK(so);
232	return -1;
233	}
234	}
235
236	#ifndef HAVE_READV
237	/*
238	* If there was no error, try and read the second time round
239	* We read again if n = 2 (ie, there's another part of the buffer)
240	* and we read as much as we could in the first read
241	* We don't test for <= 0 this time, because there legitimately
242	* might not be any more data (since the socket is non-blocking),
243	* a close will be detected on next iteration.
244	* A return of -1 wont (shouldn't) happen, since it didn't happen above
245	*/
246	if (n == 2 && nn == iov[0].iov_len)
247	{
248	int ret;
249	ret = recv(so->s, iov[1].iov_base, iov[1].iov_len, 0);
250	if (ret > 0)
251	nn += ret;
252	}
253
254	DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
255	#endif
256
257	/* Update fields */
258	sb->sb_cc += nn;
259	sb->sb_wptr += nn;
260	if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
261	sb->sb_wptr -= sb->sb_datalen;
262	SOCKET_UNLOCK(so);
263	return nn;
264	}
265
266	/*
267	* Get urgent data
268	*
269	* When the socket is created, we set it SO_OOBINLINE,
270	* so when OOB data arrives, we soread() it and everything
271	* in the send buffer is sent as urgent data
272	*/
273	void
274	sorecvoob(PNATState pData, struct socket *so)
275	{
276	struct tcpcb *tp = sototcpcb(so);
277
278	DEBUG_CALL("sorecvoob");
279	DEBUG_ARG("so = %lx", (long)so);
280
281	/*
282	* We take a guess at how much urgent data has arrived.
283	* In most situations, when urgent data arrives, the next
284	* read() should get all the urgent data. This guess will
285	* be wrong however if more data arrives just after the
286	* urgent data, or the read() doesn't return all the
287	* urgent data.
288	*/
289	soread(pData, so);
290	tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
291	tp->t_force = 1;
292	tcp_output(pData, tp);
293	tp->t_force = 0;
294	}
295
296	/*
297	* Send urgent data
298	* There's a lot duplicated code here, but...
299	*/
300	int
301	sosendoob(struct socket *so)
302	{
303	struct sbuf *sb = &so->so_rcv;
304	char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
305
306	int n, len;
307
308	DEBUG_CALL("sosendoob");
309	DEBUG_ARG("so = %lx", (long)so);
310	DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
311
312	if (so->so_urgc > sizeof(buff))
313	so->so_urgc = sizeof(buff); /* XXX */
314
315	if (sb->sb_rptr < sb->sb_wptr)
316	{
317	/* We can send it directly */
318	n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
319	so->so_urgc -= n;
320
321	DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n",
322	n, so->so_urgc));
323	}
324	else
325	{
326	/*
327	* Since there's no sendv or sendtov like writev,
328	* we must copy all data to a linear buffer then
329	* send it all
330	*/
331	len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
332	if (len > so->so_urgc)
333	len = so->so_urgc;
334	memcpy(buff, sb->sb_rptr, len);
335	so->so_urgc -= len;
336	if (so->so_urgc)
337	{
338	n = sb->sb_wptr - sb->sb_data;
339	if (n > so->so_urgc)
340	n = so->so_urgc;
341	memcpy(buff + len, sb->sb_data, n);
342	so->so_urgc -= n;
343	len += n;
344	}
345	n = send(so->s, buff, len, (MSG_OOB)); /* \|MSG_DONTWAIT)); */
346	#ifdef DEBUG
347	if (n != len)
348	DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
349	#endif
350	DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n",
351	n, so->so_urgc));
352	}
353
354	sb->sb_cc -= n;
355	sb->sb_rptr += n;
356	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
357	sb->sb_rptr -= sb->sb_datalen;
358
359	return n;
360	}
361
362	/*
363	* Write data from so_rcv to so's socket,
364	* updating all sbuf field as necessary
365	*/
366	int
367	sowrite(PNATState pData, struct socket *so)
368	{
369	int n, nn;
370	struct sbuf *sb = &so->so_rcv;
371	size_t len = sb->sb_cc;
372	struct iovec iov[2];
373
374	DEBUG_CALL("sowrite");
375	DEBUG_ARG("so = %lx", (long)so);
376	QSOCKET_LOCK(tcb);
377	SOCKET_LOCK(so);
378	QSOCKET_UNLOCK(tcb);
379	if (so->so_urgc)
380	{
381	sosendoob(so);
382	if (sb->sb_cc == 0)
383	{
384	SOCKET_UNLOCK(so);
385	return 0;
386	}
387	}
388
389	/*
390	* No need to check if there's something to write,
391	* sowrite wouldn't have been called otherwise
392	*/
393
394	len = sb->sb_cc;
395
396	iov[0].iov_base = sb->sb_rptr;
397	iov[1].iov_base = 0;
398	iov[1].iov_len = 0;
399	if (sb->sb_rptr < sb->sb_wptr)
400	{
401	iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
402	/* Should never succeed, but... */
403	if (iov[0].iov_len > len)
404	iov[0].iov_len = len;
405	n = 1;
406	}
407	else
408	{
409	iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
410	if (iov[0].iov_len > len)
411	iov[0].iov_len = len;
412	len -= iov[0].iov_len;
413	if (len)
414	{
415	iov[1].iov_base = sb->sb_data;
416	iov[1].iov_len = sb->sb_wptr - sb->sb_data;
417	if (iov[1].iov_len > len)
418	iov[1].iov_len = len;
419	n = 2;
420	}
421	else
422	n = 1;
423	}
424	/* Check if there's urgent data to send, and if so, send it */
425	#ifdef HAVE_READV
426	nn = writev(so->s, (const struct iovec *)iov, n);
427	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
428	#else
429	nn = send(so->s, iov[0].iov_base, iov[0].iov_len, 0);
430	#endif
431	/* This should never happen, but people tell me it does shrug */
432	if (nn < 0 && (errno == EAGAIN \|\| errno == EINTR \|\| errno == EWOULDBLOCK))
433	{
434	SOCKET_UNLOCK(so);
435	return 0;
436	}
437
438	if (nn < 0 \|\| (nn == 0 && iov[0].iov_len > 0))
439	{
440	DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
441	so->so_state, errno));
442	sofcantsendmore(so);
443	tcp_sockclosed(pData, sototcpcb(so));
444	SOCKET_UNLOCK(so);
445	return -1;
446	}
447
448	#ifndef HAVE_READV
449	if (n == 2 && nn == iov[0].iov_len)
450	{
451	int ret;
452	ret = send(so->s, iov[1].iov_base, iov[1].iov_len, 0);
453	if (ret > 0)
454	nn += ret;
455	}
456	DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
457	#endif
458
459	/* Update sbuf */
460	sb->sb_cc -= nn;
461	sb->sb_rptr += nn;
462	if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
463	sb->sb_rptr -= sb->sb_datalen;
464
465	/*
466	* If in DRAIN mode, and there's no more data, set
467	* it CANTSENDMORE
468	*/
469	if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
470	sofcantsendmore(so);
471
472	SOCKET_UNLOCK(so);
473	return nn;
474	}
475
476	/*
477	* recvfrom() a UDP socket
478	*/
479	void
480	sorecvfrom(PNATState pData, struct socket *so)
481	{
482	struct sockaddr_in addr;
483	socklen_t addrlen = sizeof(struct sockaddr_in);
484
485	DEBUG_CALL("sorecvfrom");
486	DEBUG_ARG("so = %lx", (long)so);
487
488	if (so->so_type == IPPROTO_ICMP)
489	{
490	/* This is a "ping" reply */
491	#ifdef RT_OS_WINDOWS
492	sorecvfrom_icmp_win(pData, so);
493	#else /* RT_OS_WINDOWS */
494	sorecvfrom_icmp_unix(pData, so);
495	#endif /* !RT_OS_WINDOWS */
496	udp_detach(pData, so);
497	}
498	else
499	{
500	/* A "normal" UDP packet */
501	struct mbuf *m;
502	struct ethhdr *eh;
503	size_t len;
504	u_long n;
505
506	QSOCKET_LOCK(udb);
507	SOCKET_LOCK(so);
508	QSOCKET_UNLOCK(udb);
509
510	if (!(m = m_get(pData)))
511	{
512	SOCKET_UNLOCK(so);
513	return;
514	}
515	/* adjust both parameters to maks M_FREEROOM calculate correct */
516	m_adj(m, if_maxlinkhdr + sizeof(struct udphdr) + sizeof(struct ip));
517
518	/*
519	* XXX Shouldn't FIONREAD packets destined for port 53,
520	* but I don't know the max packet size for DNS lookups
521	*/
522	len = M_FREEROOM(m);
523	/* if (so->so_fport != htons(53)) */
524	{
525	ioctlsocket(so->s, FIONREAD, &n);
526
527	if (n > len)
528	{
529	n = (m->m_data - m->m_dat) + m->m_len + n + 1;
530	m_inc(m, n);
531	len = M_FREEROOM(m);
532	}
533	}
534
535	m->m_len = recvfrom(so->s, m->m_data, len, 0,
536	(struct sockaddr *)&addr, &addrlen);
537	Log2((" did recvfrom %d, errno = %d-%s\n",
538	m->m_len, errno, strerror(errno)));
539	if(m->m_len < 0)
540	{
541	u_char code = ICMP_UNREACH_PORT;
542
543	if (errno == EHOSTUNREACH)
544	code = ICMP_UNREACH_HOST;
545	else if(errno == ENETUNREACH)
546	code = ICMP_UNREACH_NET;
547
548	Log2((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
549	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
550	so->so_m = NULL;
551	m_free(pData, m);
552	}
553	else
554	{
555	/*
556	* Hack: domain name lookup will be used the most for UDP,
557	* and since they'll only be used once there's no need
558	* for the 4 minute (or whatever) timeout... So we time them
559	* out much quicker (10 seconds for now...)
560	*/
561	#ifndef VBOX_WITH_SLIRP_DNS_PROXY
562	if (so->so_expire)
563	{
564	if (so->so_fport == htons(53))
565	so->so_expire = curtime + SO_EXPIREFAST;
566	else
567	so->so_expire = curtime + SO_EXPIRE;
568	}
569	#else
570	if (so->so_expire)
571	{
572	if (so->so_fport != htons(53))
573	so->so_expire = curtime + SO_EXPIRE;
574	}
575	/*
576	* last argument should be changed if Slirp will inject IP attributes
577	* Note: Here we can't check if dnsproxy's sent initial request
578	*/
579	if (so->so_fport == htons(53))
580	dnsproxy_answer(pData, so, m);
581	#endif
582
583	#if 0
584	if (m->m_len == len)
585	{
586	m_inc(m, MINCSIZE);
587	m->m_len = 0;
588	}
589	#endif
590
591	/*
592	* If this packet was destined for CTL_ADDR,
593	* make it look like that's where it came from, done by udp_output
594	*/
595	udp_output(pData, so, m, &addr);
596	SOCKET_UNLOCK(so);
597	} /* rx error */
598	} /* if ping packet */
599	}
600
601	/*
602	* sendto() a socket
603	*/
604	int
605	sosendto(PNATState pData, struct socket so, struct mbuf m)
606	{
607	int ret;
608	struct sockaddr_in addr;
609	#if 0
610	struct sockaddr_in host_addr;
611	#endif
612
613	DEBUG_CALL("sosendto");
614	DEBUG_ARG("so = %lx", (long)so);
615	DEBUG_ARG("m = %lx", (long)m);
616
617	addr.sin_family = AF_INET;
618	if ((so->so_faddr.s_addr & htonl(pData->netmask)) == special_addr.s_addr)
619	{
620	/* It's an alias */
621	uint32_t last_byte = ntohl(so->so_faddr.s_addr) & ~pData->netmask;
622	switch(last_byte)
623	{
624	#if 0
625	/* handle this case at 'default:' */
626	case CTL_BROADCAST:
627	addr.sin_addr.s_addr = INADDR_BROADCAST;
628	/* Send the packet to host to fully emulate broadcast */
629	/** @todo r=klaus: on Linux host this causes the host to receive
630	* the packet twice for some reason. And I cannot find any place
631	* in the man pages which states that sending a broadcast does not
632	* reach the host itself. */
633	host_addr.sin_family = AF_INET;
634	host_addr.sin_port = so->so_fport;
635	host_addr.sin_addr = our_addr;
636	sendto(so->s, m->m_data, m->m_len, 0,
637	(struct sockaddr *)&host_addr, sizeof (struct sockaddr));
638	break;
639	#endif
640	case CTL_DNS:
641	#ifndef VBOX_WITH_MULTI_DNS
642	if (!get_dns_addr(pData, &dns_addr))
643	addr.sin_addr = dns_addr;
644	else
645	addr.sin_addr = loopback_addr;
646	break;
647	#endif
648	case CTL_ALIAS:
649	default:
650	if (last_byte == ~pData->netmask)
651	addr.sin_addr.s_addr = INADDR_BROADCAST;
652	else
653	addr.sin_addr = loopback_addr;
654	break;
655	}
656	}
657	else
658	addr.sin_addr = so->so_faddr;
659	addr.sin_port = so->so_fport;
660
661	DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n",
662	ntohs(addr.sin_port), inet_ntoa(addr.sin_addr)));
663
664	/* Don't care what port we get */
665	ret = sendto(so->s, m->m_data, m->m_len, 0,
666	(struct sockaddr *)&addr, sizeof (struct sockaddr));
667	if (ret < 0)
668	{
669	LogRel(("UDP: sendto fails (%s)\n", strerror(errno)));
670	return -1;
671	}
672
673	/*
674	* Kill the socket if there's no reply in 4 minutes,
675	* but only if it's an expirable socket
676	*/
677	if (so->so_expire)
678	so->so_expire = curtime + SO_EXPIRE;
679	so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
680	return 0;
681	}
682
683	/*
684	* XXX This should really be tcp_listen
685	*/
686	struct socket *
687	solisten(PNATState pData, u_int port, u_int32_t laddr, u_int lport, int flags)
688	{
689	struct sockaddr_in addr;
690	struct socket *so;
691	socklen_t addrlen = sizeof(addr);
692	int s, opt = 1;
693	int status;
694
695	DEBUG_CALL("solisten");
696	DEBUG_ARG("port = %d", port);
697	DEBUG_ARG("laddr = %x", laddr);
698	DEBUG_ARG("lport = %d", lport);
699	DEBUG_ARG("flags = %x", flags);
700
701	if ((so = socreate()) == NULL)
702	{
703	/* RTMemFree(so); Not sofree() ??? free(NULL) == NOP */
704	return NULL;
705	}
706
707	/* Don't tcp_attach... we don't need so_snd nor so_rcv */
708	if ((so->so_tcpcb = tcp_newtcpcb(pData, so)) == NULL)
709	{
710	RTMemFree(so);
711	return NULL;
712	}
713
714	SOCKET_LOCK_CREATE(so);
715	SOCKET_LOCK(so);
716	QSOCKET_LOCK(tcb);
717	insque(pData, so,&tcb);
718	NSOCK_INC();
719	QSOCKET_UNLOCK(tcb);
720
721	/*
722	* SS_FACCEPTONCE sockets must time out.
723	*/
724	if (flags & SS_FACCEPTONCE)
725	so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
726
727	so->so_state = (SS_FACCEPTCONN\|flags);
728	so->so_lport = lport; /* Kept in network format */
729	so->so_laddr.s_addr = laddr; /* Ditto */
730
731	addr.sin_family = AF_INET;
732	addr.sin_addr.s_addr = INADDR_ANY;
733	addr.sin_port = port;
734
735	if ( ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
736	\|\| (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,(char *)&opt, sizeof(int)) < 0)
737	\|\| (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0)
738	\|\| (listen(s, 1) < 0))
739	{
740	#ifdef RT_OS_WINDOWS
741	int tmperrno = WSAGetLastError(); /* Don't clobber the real reason we failed */
742	closesocket(s);
743	QSOCKET_LOCK(tcb);
744	sofree(pData, so);
745	QSOCKET_UNLOCK(tcb);
746	/* Restore the real errno */
747	WSASetLastError(tmperrno);
748	#else
749	int tmperrno = errno; /* Don't clobber the real reason we failed */
750	close(s);
751	QSOCKET_LOCK(tcb);
752	sofree(pData, so);
753	QSOCKET_UNLOCK(tcb);
754	/* Restore the real errno */
755	errno = tmperrno;
756	#endif
757	return NULL;
758	}
759	fd_nonblock(s);
760	setsockopt(s, SOL_SOCKET, SO_OOBINLINE,(char *)&opt, sizeof(int));
761
762	getsockname(s,(struct sockaddr *)&addr,&addrlen);
763	so->so_fport = addr.sin_port;
764	/* set socket buffers */
765	opt = 64 * _1K;
766	status = setsockopt(s, SOL_SOCKET, SO_RCVBUF, (char *)&opt, sizeof(int));
767	if (status < 0)
768	{
769	LogRel(("NAT: Error(%d) while setting RCV capacity to (%d)\n", errno, opt));
770	goto no_sockopt;
771	}
772	opt = 64 * _1K;
773	status = setsockopt(s, SOL_SOCKET, SO_SNDBUF, (char *)&opt, sizeof(int));
774	if (status < 0)
775	{
776	LogRel(("NAT: Error(%d) while setting SND capacity to (%d)\n", errno, opt));
777	goto no_sockopt;
778	}
779	no_sockopt:
780	if (addr.sin_addr.s_addr == 0 \|\| addr.sin_addr.s_addr == loopback_addr.s_addr)
781	so->so_faddr = alias_addr;
782	else
783	so->so_faddr = addr.sin_addr;
784
785	so->s = s;
786	SOCKET_UNLOCK(so);
787	return so;
788	}
789
790	/*
791	* Data is available in so_rcv
792	* Just write() the data to the socket
793	* XXX not yet...
794	*/
795	void
796	sorwakeup(struct socket *so)
797	{
798	#if 0
799	sowrite(so);
800	FD_CLR(so->s,&writefds);
801	#endif
802	}
803
804	/*
805	* Data has been freed in so_snd
806	* We have room for a read() if we want to
807	* For now, don't read, it'll be done in the main loop
808	*/
809	void
810	sowwakeup(struct socket *so)
811	{
812	}
813
814	/*
815	* Various session state calls
816	* XXX Should be #define's
817	* The socket state stuff needs work, these often get call 2 or 3
818	* times each when only 1 was needed
819	*/
820	void
821	soisfconnecting(struct socket *so)
822	{
823	so->so_state &= ~(SS_NOFDREF\|SS_ISFCONNECTED\|SS_FCANTRCVMORE\|
824	SS_FCANTSENDMORE\|SS_FWDRAIN);
825	so->so_state \|= SS_ISFCONNECTING; /* Clobber other states */
826	}
827
828	void
829	soisfconnected(struct socket *so)
830	{
831	so->so_state &= ~(SS_ISFCONNECTING\|SS_FWDRAIN\|SS_NOFDREF);
832	so->so_state \|= SS_ISFCONNECTED; /* Clobber other states */
833	}
834
835	void
836	sofcantrcvmore(struct socket *so)
837	{
838	if ((so->so_state & SS_NOFDREF) == 0)
839	{
840	shutdown(so->s, 0);
841	}
842	so->so_state &= ~(SS_ISFCONNECTING);
843	if (so->so_state & SS_FCANTSENDMORE)
844	so->so_state = SS_NOFDREF; /* Don't select it */
845	/* XXX close() here as well? */
846	else
847	so->so_state \|= SS_FCANTRCVMORE;
848	}
849
850	void
851	sofcantsendmore(struct socket *so)
852	{
853	if ((so->so_state & SS_NOFDREF) == 0)
854	shutdown(so->s, 1); /* send FIN to fhost */
855
856	so->so_state &= ~(SS_ISFCONNECTING);
857	if (so->so_state & SS_FCANTRCVMORE)
858	so->so_state = SS_NOFDREF; /* as above */
859	else
860	so->so_state \|= SS_FCANTSENDMORE;
861	}
862
863	void
864	soisfdisconnected(struct socket *so)
865	{
866	#if 0
867	so->so_state &= ~(SS_ISFCONNECTING\|SS_ISFCONNECTED);
868	close(so->s);
869	so->so_state = SS_ISFDISCONNECTED;
870	/*
871	* XXX Do nothing ... ?
872	*/
873	#endif
874	}
875
876	/*
877	* Set write drain mode
878	* Set CANTSENDMORE once all data has been write()n
879	*/
880	void
881	sofwdrain(struct socket *so)
882	{
883	if (so->so_rcv.sb_cc)
884	so->so_state \|= SS_FWDRAIN;
885	else
886	sofcantsendmore(so);
887	}
888
889	static void
890	send_icmp_to_guest(PNATState pData, char buff, size_t len, struct socket so, const struct sockaddr_in *addr)
891	{
892	struct ip *ip;
893	uint32_t dst, src;
894	char ip_copy[256];
895	struct icmp *icp;
896	int old_ip_len = 0;
897	int hlen, original_hlen = 0;
898	struct mbuf *m;
899	struct icmp_msg *icm;
900	uint8_t proto;
901	int type = 0;
902
903	ip = (struct ip *)buff;
904	hlen = (ip->ip_hl << 2);
905	icp = (struct icmp )((char )ip + hlen);
906
907	Log(("ICMP:received msg(t:%d, c:%d)\n", icp->icmp_type, icp->icmp_code));
908	if ( icp->icmp_type != ICMP_ECHOREPLY
909	&& icp->icmp_type != ICMP_TIMXCEED
910	&& icp->icmp_type != ICMP_UNREACH)
911	{
912	return;
913	}
914
915	type = icp->icmp_type;
916	if ( type == ICMP_TIMXCEED
917	\|\| type == ICMP_UNREACH)
918	{
919	ip = &icp->icmp_ip;
920	DO_ALIAS(&ip->ip_dst);
921	}
922	else
923	{
924	DO_ALIAS(&ip->ip_src);
925	}
926
927	icm = icmp_find_original_mbuf(pData, ip);
928
929	if (icm == NULL)
930	{
931	Log(("NAT: Can't find the corresponding packet for the received ICMP\n"));
932	return;
933	}
934
935	m = icm->im_m;
936	Assert(m != NULL);
937
938	src = addr->sin_addr.s_addr;
939
940	ip = mtod(m, struct ip *);
941	proto = ip->ip_p;
942	/* Now ip is pointing on header we've sent from guest */
943	if ( icp->icmp_type == ICMP_TIMXCEED
944	\|\| icp->icmp_type == ICMP_UNREACH)
945	{
946	old_ip_len = (ip->ip_hl << 2) + 64;
947	if (old_ip_len > sizeof(ip_copy))
948	old_ip_len = sizeof(ip_copy);
949	memcpy(ip_copy, ip, old_ip_len);
950	}
951
952	/* source address from original IP packet*/
953	dst = ip->ip_src.s_addr;
954
955	/* overide ther tail of old packet */
956	ip = mtod(m, struct ip ); / ip is from mbuf we've overrided */
957	original_hlen = ip->ip_hl << 2;
958	/* saves original ip header and options */
959	memcpy(m->m_data + original_hlen, buff + hlen, len - hlen);
960	m->m_len = len - hlen + original_hlen;
961	ip->ip_len = m->m_len;
962	ip->ip_p = IPPROTO_ICMP; /* the original package could be whatever, but we're response via ICMP*/
963
964	icp = (struct icmp )((char )ip + (ip->ip_hl << 2));
965	type = icp->icmp_type;
966	if ( type == ICMP_TIMXCEED
967	\|\| type == ICMP_UNREACH)
968	{
969	/* according RFC 793 error messages required copy of initial IP header + 64 bit */
970	memcpy(&icp->icmp_ip, ip_copy, old_ip_len);
971	ip->ip_tos = ((ip->ip_tos & 0x1E) \| 0xC0); /* high priority for errors */
972	}
973
974	ip->ip_src.s_addr = src;
975	ip->ip_dst.s_addr = dst;
976	icmp_reflect(pData, m);
977	LIST_REMOVE(icm, im_list);
978	/* Don't call m_free here*/
979
980	if ( type == ICMP_TIMXCEED
981	\|\| type == ICMP_UNREACH)
982	{
983	icm->im_so->so_m = NULL;
984	switch (proto)
985	{
986	case IPPROTO_UDP:
987	/XXX: so->so_m already freed so we shouldn't call sofree /
988	udp_detach(pData, icm->im_so);
989	break;
990	case IPPROTO_TCP:
991	/close tcp should be here /
992	break;
993	default:
994	/* do nothing */
995	break;
996	}
997	}
998	RTMemFree(icm);
999	}
1000
1001	#ifdef RT_OS_WINDOWS
1002	static void
1003	sorecvfrom_icmp_win(PNATState pData, struct socket *so)
1004	{
1005	int len;
1006	int i;
1007	struct ip *ip;
1008	struct mbuf *m;
1009	struct icmp *icp;
1010	struct icmp_msg *icm;
1011	struct ip ip_broken; / ICMP returns header + 64 bit of packet */
1012	uint32_t src;
1013	ICMP_ECHO_REPLY *icr;
1014	int hlen = 0;
1015	int data_len = 0;
1016	int nbytes = 0;
1017	u_char code = ~0;
1018
1019	len = pData->pfIcmpParseReplies(pData->pvIcmpBuffer, pData->szIcmpBuffer);
1020	if (len < 0)
1021	{
1022	LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError()));
1023	return;
1024	}
1025	if (len == 0)
1026	return; /* no error */
1027
1028	icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer;
1029	for (i = 0; i < len; ++i)
1030	{
1031	switch(icr[i].Status)
1032	{
1033	case IP_DEST_HOST_UNREACHABLE:
1034	code = (code != ~0 ? code : ICMP_UNREACH_HOST);
1035	case IP_DEST_NET_UNREACHABLE:
1036	code = (code != ~0 ? code : ICMP_UNREACH_NET);
1037	case IP_DEST_PROT_UNREACHABLE:
1038	code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL);
1039	/* UNREACH error inject here */
1040	case IP_DEST_PORT_UNREACHABLE:
1041	code = (code != ~0 ? code : ICMP_UNREACH_PORT);
1042	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!");
1043	so->so_m = NULL;
1044	break;
1045	case IP_SUCCESS: /* echo replied */
1046	m = m_get(pData);
1047	m->m_data += if_maxlinkhdr;
1048	ip = mtod(m, struct ip *);
1049	ip->ip_src.s_addr = icr[i].Address;
1050	DO_ALIAS(&ip->ip_src);
1051	ip->ip_p = IPPROTO_ICMP;
1052	ip->ip_dst.s_addr = so->so_laddr.s_addr; /XXX: still the hack/
1053	data_len = sizeof(struct ip);
1054	ip->ip_hl = data_len >> 2; /* requiered for icmp_reflect, no IP options */
1055	ip->ip_ttl = icr[i].Options.Ttl;
1056
1057	icp = (struct icmp )&ip[1]; / no options */
1058	icp->icmp_type = ICMP_ECHOREPLY;
1059	icp->icmp_code = 0;
1060	icp->icmp_id = so->so_icmp_id;
1061	icp->icmp_seq = so->so_icmp_seq;
1062
1063	data_len += ICMP_MINLEN;
1064
1065	nbytes = (data_len + icr[i].DataSize > m->m_size? m->m_size - data_len: icr[i].DataSize);
1066	memcpy(icp->icmp_data, icr[i].Data, nbytes);
1067
1068	data_len += icr[i].DataSize;
1069
1070	ip->ip_len = data_len;
1071	m->m_len = ip->ip_len;
1072
1073	icmp_reflect(pData, m);
1074	break;
1075	case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */
1076
1077	ip_broken = icr[i].Data;
1078	icm = icmp_find_original_mbuf(pData, ip_broken);
1079	if (icm == NULL) {
1080	Log(("ICMP: can't find original package (first double word %x)\n", (uint32_t )ip_broken));
1081	return;
1082	}
1083	m = icm->im_m;
1084	ip = mtod(m, struct ip *);
1085	ip->ip_ttl = icr[i].Options.Ttl;
1086	src = ip->ip_src.s_addr;
1087	ip->ip_dst.s_addr = src;
1088	ip->ip_dst.s_addr = icr[i].Address;
1089
1090	hlen = (ip->ip_hl << 2);
1091	icp = (struct icmp )((char )ip + hlen);
1092	ip_broken->ip_src.s_addr = src; /it packet sent from host not from guest/
1093	data_len = (ip_broken->ip_hl << 2) + 64;
1094
1095	nbytes =(hlen + ICMP_MINLEN + data_len > m->m_size? m->m_size - (hlen + ICMP_MINLEN): data_len);
1096	memcpy(icp->icmp_data, ip_broken, nbytes);
1097	icmp_reflect(pData, m);
1098	break;
1099	default:
1100	Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address));
1101	break;
1102	}
1103	}
1104	}
1105	#else /* RT_OS_WINDOWS */
1106	static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so)
1107	{
1108	struct sockaddr_in addr;
1109	socklen_t addrlen = sizeof(struct sockaddr_in);
1110	char buff[1500];
1111	int len;
1112	len = recvfrom(so->s, buff, 1500, 0,
1113	(struct sockaddr *)&addr, &addrlen);
1114	/* XXX Check if reply is "correct"? */
1115
1116	if (len == -1 \|\| len == 0)
1117	{
1118	u_char code = ICMP_UNREACH_PORT;
1119
1120	if (errno == EHOSTUNREACH)
1121	code = ICMP_UNREACH_HOST;
1122	else if(errno == ENETUNREACH)
1123	code = ICMP_UNREACH_NET;
1124
1125	DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
1126	errno, strerror(errno)));
1127	icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
1128	so->so_m = NULL;
1129	}
1130	else
1131	{
1132	send_icmp_to_guest(pData, buff, len, so, &addr);
1133	}
1134	}
1135	#endif /* !RT_OS_WINDOWS */

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/Devices/Network/slirp/socket.c@ 20297

Download in other formats: